SourceMay 30, 20262026-05-304 minVerified
PRO-CUA Shifts Browser Agents Toward Step-Level Rewards
PRO-CUA trains browser agents with step-level process rewards instead of trajectory-only signals, targeting credit assignment.
CommunityMay 30, 20262026-05-305 minVerified
Reselling AI Access: Subscriptions, APIs, and Policy Limits
Explains how subscription and API billing differ, and why reselling AI access raises policy, security, and operational risks.
SourceMay 30, 20262026-05-305 minVerified
Streaming Audits for Group-Conditioned News Framing Bias
A streaming evaluation approach that tracks how LLM news framing shifts across groups as events, models, and systems change.
SourceMay 30, 20262026-05-304 minVerified
Student-Centered Data Selection for Reasoning Distillation
DMC suggests student-model compatibility, not just data quality, may matter more for reasoning distillation.
CommunityMay 29, 20262026-05-294 minVerified
AI as a Strategic Assistant for Mathematics Work
Why AI may matter more as a long-horizon task worker and strategic assistant in mathematics than as an answer generator.
SourceMay 29, 20262026-05-294 minVerified
Comparing Agentic AI for End-to-End Gravitational Wave Pipelines
A head-to-head test of Claude Code and Codex running an end-to-end gravitational wave analysis pipeline autonomously.
SourceMay 29, 20262026-05-294 minVerified
Designing Triadic LLM Collaboration for K-12 Writing
An arXiv study examines teacher-student-model collaboration and control frameworks for LLM use in K-12 writing.
SourceMay 29, 20262026-05-295 minVerified
Governing Technical Debt in Agentic AI Systems
Why agentic AI failures create governance and operational control risks beyond model accuracy alone.
SourceMay 29, 20262026-05-294 minVerified
SCDBench Shifts Smart Contract Decompilation Toward Semantic Evaluation
SCDBench argues smart contract decompilation should be judged by semantic equivalence, not just source-like Solidity.
SourceMay 29, 20262026-05-294 minVerified
Streaming Synthetic Data Learning Across Sequential Tasks
Examines synthetic data generation as a streaming learning problem, focusing on transfer, forgetting, and feedback loops.
SourceMay 29, 20262026-05-295 minVerified
TaxDistill Rethinks Metagenome Classification Beyond Model Size
TaxDistill argues pretraining data composition and distilled genome representations matter more than model size.
SourceMay 29, 20262026-05-294 minVerified
Time Structure Limits in Tokenized Time Series LLMs
This study argues tokenized time series LLMs lose continuity and order, and proposes COM constraints to preserve temporal structure.
SourceMay 29, 20262026-05-294 minVerified
Trace-Level Synthesis Challenges Consensus in Multi-Agent Reasoning
A look at a paper arguing that aggregating full reasoning traces can outperform answer-only consensus in multi-agent systems.
SourceMay 29, 20262026-05-295 minVerified
VitalAgent for Long-Term ECG and PPG Reasoning
VitalAgent proposes an agent architecture for long-term ECG and PPG streams with reasoning, memory, and proactive monitoring.
SourceMay 28, 20262026-05-284 minVerified
Human-AI Collaboration in Scientific Replicability Assessment
Examines human-AI collaboration for replicability prediction, balancing speed and consistency against bias, accountability, and privacy risks.
SourceMay 28, 20262026-05-285 minVerified
MOV-Bench Reveals Gaps in Multi-Hop Video Reasoning
MOV-Bench highlights evaluation gaps in multi-hop audio-visual reasoning and shows consistent gains from agentic search.
SourceMay 28, 20262026-05-284 minVerified
Reassessing Offline RL for Code Generation Post-Training
Examines whether offline RL can cut online RL costs in code generation post-training without sacrificing practical quality.
SourceMay 28, 20262026-05-284 minVerified
Turning Papers Into Benchmarks With Agentic Reproduction Workflows
How under-specified applied ML papers can become executable benchmarks through agentic workflows and slot-based reporting.
SourceMay 23, 20262026-05-234 minVerified
Policy Layers for Governing Generalist LLM Agents Safely
How policy-as-code layers can govern generalist LLM agents by controlling tool use, approvals, and data exposure.
SourceMay 21, 20262026-05-215 minVerified
Multi-Model LLM Scheduling Under Offloading And Preemption Costs
Examines how offloading and preemption affect multi-model LLM serving under GPU memory limits and model-specific costs.
SourceMay 20, 20262026-05-204 minVerified
COBALT Rethinks Robot Learning Through Smartphone Teleoperation Data
COBALT proposes smartphone and cloud teleoperation to reduce data collection bottlenecks in robot imitation learning.
SourceMay 20, 20262026-05-204 minVerified
Limits of Handwritten Math Grading With Vision LLMs
In handwritten math grading, process understanding matters more than OCR, requiring rubric-based review and human checks.
SourceMay 20, 20262026-05-204 minVerified
Multi-Image Jailbreaks Expose Multimodal LLM Safety Gaps
Multi-image prompts can bypass single-image filters, exposing structural safety gaps in multimodal LLM defenses.
SourceApr 3, 20262026-04-033 minVerified
Wrapping Florence-2 for ROS 2 Robotic Integration
A case of wrapping Florence-2 with ROS 2 topics, services, and actions for local inference and reproducible integration.